Give your coding agent a map of the repository, not a pile of search results.

codegraph is a local CLI and TypeScript library that turns a source tree into a resolved map of files, symbols, references, and dependencies. Ask where an implementation lives, how components connect, what a change can break, or which tests are relevant, then get bounded source evidence and copyable next steps.

codegraph is not an AST, a compiler, or a linter. People can use codegraph review and codegraph deps to inspect changes and dependencies without an agent.

Without structural context, an agent burns early turns listing directories, guessing search terms, opening candidate files, and reconstructing relationships. codegraph does that discovery once so the context window can stay focused on the problem.

With Node.js 22.16 or newer, install from npm:

npm install -g @lzehrung/codegraph
codegraph doctor
codegraph install
codegraph explore "build review report" --root .

No Node.js or npm? The standalone installers download a self-contained bundle with Node.js, the CLI, and the matching native runtime:

irm https://github.com/lzehrung/codegraph/releases/latest/download/install.ps1 | iex

curl -fsSL https://github.com/lzehrung/codegraph/releases/latest/download/install.sh | sh

Use codegraph alongside text search and compilers: text search finds exact strings, compilers prove language behavior, and codegraph fills in the cross-file repository map between them. See Installation for npm, standalone, and source-checkout paths.

• Changelog
• What you can do
• Try it
• A useful first five minutes
• Visualize a graph
• What the output looks like
• Why codegraph
• Why not just grep or an LSP?
• Agent setup
• Language support
• Using as a library
• How it works
• Limits and tradeoffs
• Documentation
  • CLI
  • Publishing
• Development

CLI output is readable by default. Use --json for structured fields and omission counts.

See Installation for npm, standalone, and source-checkout steps.

Do not begin by generating every possible report. Start with the question you actually have.

# Map the repository before selecting a target
codegraph orient --root . --budget small

# Search a concrete term (the query is lexical: use the words that appear in code)
codegraph search "build review report" --json

# explore: search, then bundle the top hits with source, dependents, and candidate tests
codegraph explore "build review report" --root .

# Follow a returned target
codegraph explain src/review.ts
codegraph deps src/review.ts --json
codegraph refs src/review.ts:215:23

# Compact reviewer handoff for staged and unstaged tracked changes
codegraph review

# Broader blast-radius map when the summary needs expansion
codegraph impact --base HEAD --head WORKTREE

# Deterministic affected-test paths for focused validation
codegraph affected --base HEAD --head WORKTREE --quiet

Use --head STAGED to compare HEAD with the index, or use refs such as --base origin/main --head HEAD for a branch review.

codegraph inspect ./src --limit 20
codegraph cycles --sort priority
codegraph unresolved
codegraph apisurface
codegraph duplicates ./src --min-confidence medium --limit 20
codegraph drift ./src --base origin/main --head HEAD --graph-edges summary --public-api removals

# Validate local Markdown links offline (exit 1 on broken links)
codegraph links --root .

codegraph graph --root . ./src --json --output codegraph.json
codegraph graph --root . ./src --mermaid --output graph.mmd
codegraph graph --root . ./src --dot --output graph.dot
codegraph graph --root . ./src --sqlite codegraph.sqlite

The packaged viewer is a human-facing graph UI; agents should use graph JSON, SQLite, MCP, or --json instead. Its command is codegraph viewer [--root <root>] [--graph <root-confined-json>] [--host <host>] [--port <0-65535>] [--open] [--print-url]; the root defaults to the current directory.

codegraph viewer --root . --open
codegraph viewer --root . --graph codegraph-out/graph.json --open
codegraph viewer --root . --port 4173 --print-url

The default host is 127.0.0.1 and the default port is 4173. Without --graph, each UI load or reload builds a current graph projection through the automatically validated .codegraph/cache/index-v1 index; init, index, and an exported JSON file are not prerequisites. An explicit --graph serves that root-confined snapshot through the same /graph.json route, while --print-url only prints the deterministic URL and exits.

The UI loads Sigma, Graphology, and ForceAtlas2 from bundled src/viewer/vendor/ assets, so the viewer stays offline and self-contained once codegraph is installed.

Because ranking and counts change with the working tree, this abbreviated explore excerpt shows the stable response structure rather than snapshot-specific totals:

Anchors
- buildReviewReport [symbol] src/review.ts
- src/cli/help.ts:1 [chunk] src/cli/help.ts
- ReviewPreset [symbol] src/review.ts

Relevant source
- buildReviewReport is defined in src/review.ts.
- References, dependencies, and dependents are summarized here.

Blast radius
- src/review.ts: src/index.ts, src/cli/review.ts, src/mcp/server.ts, ...

Candidate tests
- tests/agent-explain.test.ts
- tests/agent-explore.test.ts
- tests/agent-packet.test.ts

Follow-ups
- codegraph file src/review.ts
- codegraph refs src/review.ts:215:23

Limits
- anchors, packets, paths, blast radius, reverse dependencies, and candidate tests

Recommended next: codegraph file src/review.ts

Real output includes counts, copyable follow-ups, and the limits that applied; --json adds the omission counts.

A worktree review is optimized for a different job:

Review Summary
==============
Status: ok
Files changed: 5
Symbols changed: 22
Candidate tests: 1 (high: 1, medium: 0, low: 0)
Risk: high (80)
Signals: exported-symbols-changed, many-symbols-changed

Structured output carries the underlying changed files, symbols, graph edges, reasons, diagnostics, snippets, and candidate-test confidence.

explore runs one search and bundles bounded context around the top hits: source, dependency paths, dependents, candidate tests, and next commands. It is a reading starting point, only as good as the search terms. MCP omits source by default.

Results include source paths, symbol ranges, stable handles, rank reasons, graph relationships, confidence, and omission counts. An agent can inspect why something ranked, jump to the definition or references, and continue from an exact target instead of treating a fuzzy match as an answer.

Definitions, references, and call edges come from scopes, imports, packages, and receivers proven from source, not from name matching alone. When codegraph cannot prove a target, it returns not_found or reports referenceCoverage as partial with a reason instead of guessing. An agent can act on a complete result and fall back to text search or a compiler for the rest; complete covers every candidate known to the index, not unknown dynamic loading. See the refs coverage contract and the language parity matrix for per-language support and tested limits.

Search, navigation, dependency analysis, impact, and review reuse the same graph and semantic index. A target found during discovery can flow directly into explain, refs, deps, impact analysis, and candidate-test selection.

One repository model can include source code, SQL, workspace packages, documentation links, stylesheets, templates, and single-file components. Capability claims stay language-specific, so graph support is not presented as full compiler or language-server parity.

codegraph runs locally as a CLI, library, or MCP server. Humans get readable output; agents and programs can keep structured JSON, stable handles, warm sessions, SQLite data, or graph exports without parsing display text.

codegraph complements both.

• Use text search for exact strings, logs, config keys, and prose.
• Use a compiler or language server when you need compiler-grade type analysis, overload resolution, dynamic dispatch, or editor refactors.
• Use codegraph when the question crosses files, languages, dependency edges, a git diff, or an agent context boundary.

The distinction is evidence shape, not a claim that one tool replaces the others.

Run codegraph install on an interactive terminal to detect supported clients, preview the changes, and confirm once. Use --all when you want the full current catalog without detection:

codegraph install
codegraph install --target codex,claude --dry-run
codegraph install --target codex,claude --yes
codegraph install --all --dry-run
codegraph install --all --yes
codegraph install --print-config codex

Supported target ids are codex, claude, cursor, gemini, opencode, omp, kilo, and agents. Interactive writes default to no; noninteractive writes require --yes.

For a skill without MCP configuration:

codegraph skill install --agent codex
codegraph skill install --agent claude
codegraph skill install --agent cursor

See Agent workflows for exploration strategy, warm sessions, streaming, review loops, and tool wrappers. Default stdio MCP clients share one compatible project owner's index, workers and watcher. Use codegraph server status --root . --json and codegraph server stop --root . to inspect or stop it; use an explicit workspace when launching from a directory of repositories. See MCP for server and client configuration.

Shared source-language indexing and navigation: JavaScript, TypeScript, Python, PHP, Go, Java, C#, Ruby, Rust, Kotlin, Swift, Zig, C, and C++.

SQL: statement chunking, object symbols, common DDL/DML and CTE facts, SQL-to-SQL edges, and object-level navigation. codegraph does not claim column-definition resolution.

Terraform: module-directory symbols, dependency edges to local module sources, and navigation for var, local, module, data source, and resource references, including module outputs and call arguments.

Graph-first formats: HTML, Astro, Handlebars, Markdown, MDX, reStructuredText, AsciiDoc, CSS, SCSS, and Less have narrower graph or chunking support.

Single-file components: Vue and Svelte script blocks participate in dependency graphs and chunking; semantic navigation is narrower.

See Language parity for the capability matrix and Scenario catalog for the fixtures behind those claims.

Install @lzehrung/codegraph-core when you need the library without the CLI, MCP server, installer, or viewer.

import { buildProjectIndex, getHotspots } from "@lzehrung/codegraph-core";

const index = await buildProjectIndex(process.cwd());
console.log(getHotspots(index.graph, { limit: 10 }));

See the Library API reference for navigation, review, agent sessions, and artifacts.

codegraph follows a single analysis pipeline:

1. Discover supported files under the selected project and include roots.
2. Parse source languages with Tree-sitter and extract imports, exports, definitions, bindings, and scopes.
3. Resolve module specifiers to project files or explicit external nodes.
4. Build forward and reverse dependency indexes plus a semantic symbol index.
5. Reuse those indexes for navigation, exploration, impact, review, and exports.

Disk caching avoids repository-wide source reads on exact warm text-search hits. How it works covers caching, native failure reporting, and performance choices.

The honest boundaries matter:

• codegraph is not a compiler or type checker. Reflection, generated code, macros, overload behavior, and dynamic dispatch can be missed. When a result would need expression type inference, overload ranking beyond arity, or build-system membership, codegraph reports not_found or partial coverage rather than a guessed target.
• Precise navigation depends on successful native parsing and language queries. A missing native addon is an error; run codegraph doctor to diagnose it. Files over the native size limit are skipped with a sourceTooLarge report, and failed native queries are reported as queryFailure without regex recovery. Document and stylesheet graph-only extraction remains supported.
• Call-compatibility findings are conservative review leads, not compiler diagnostics.
• Duplicate matches and candidate tests are ranked leads that still require human or agent judgment.
• The checked explore benchmark is a bounded evidence-retrieval benchmark, not a universal performance claim.
• The fixture test matrix and language parity docs show what codegraph has actually been tested against by language and operation; absence there means untested, not guaranteed.

Mitigations are explicit too:

• The repository keeps real-language fixtures and native parity suites for definitions, references, dependencies, chunking, and MCP.
• A generated fixture test matrix shows real tests/languages/*.test.ts status and test counts per language, with no hand-authored goldens involved.
• Structured output keeps freshness, confidence, and omission counts visible instead of pretending unsupported cases were resolved.

Run codegraph doctor to confirm the active runtime. Use --report on graph, index, search, inspect, or review commands when backend and cache behavior need to be auditable.

npm install
npm run build
npm run check

Use the narrowest relevant test while iterating. npm run check is the pre-commit baseline for formatting, lint, build, and tests; native workspace changes also require npm run build:native and npm run test:native.

codegraph is MIT licensed.